beta.div: Beta diversity computed as Var(Y)

Description Usage Arguments Details Value Author(s) References Examples

View source: R/beta.div.R


Compute estimates of total beta diversity as the total variance in a community data matrix Y, as well as derived SCBD and LCBD statistics, for 19 dissimilarity coefficients or the raw data table. Computing beta diversity as Var(Y) for raw, untransformed community composition data is not recommended. Tests of significance of the LCBD indices are also produced.


beta.div(Y, method = "hellinger", sqrt.D = FALSE, samp = TRUE,
  nperm = 999, adj = TRUE, save.D = FALSE, clock = FALSE)



Community composition data. The object class can be either data.frame or matrix.


One of the 19 dissimilarity coefficients available in the function: "hellinger", "chord", "log.chord", "chisquare", "profiles", "percentdiff", "ruzicka", "divergence", "canberra", "whittaker", "wishart", "kulczynski", "jaccard", "sorensen", "ochiai", "ab.jaccard", "ab.sorensen", "ab.ochiai", "ab.simpson", "euclidean". See Details. Names can be abbreviated to a non-ambiguous set of first letters. Default: method="hellinger".


If sqrt.D=TRUE, the dissimilarities in matrix D are square-rooted before computation of SStotal, BDtotal and LCBD. This transformation may be useful for methods "manhattan", "whittaker", "divergence", "canberra", "percentdiff", "ruzicka", "wishart" since square-root transformation of the dissimilarities makes these D matrices Euclidean.

  • Note 1 <e2><80><93> Euclideanarity is useful for ordination by principal coordinate analysis; lack of this property does not adversely affect SStotal, BDtotal and LCBD.

  • Note 2 <e2><80><93> The logical value given to parameter sqrt.D has no incidence on calculations through methods "euclidean", "profiles", "hellinger", "log.chord", "chord", "chisquare" since no D matrix is computed in those cases.

  • Note 3 <e2><80><93> For methods "jaccard", "sorensen", "ochiai", that function produces the dissimilarity matrix in the form sqrt(D), which is Euclidean.


If samp=TRUE, the abundance-based distances (ab.jaccard, ab.sorensen, ab.ochiai, ab.simpson) are computed for sample data. If samp=FALSE, they are computed for true population data.


Number of permutations for the tests of significance of LCBD indices.


Compute adjusted p-values using the Holm method. Default: adj=TRUE


If save.D=TRUE, the distance matrix will appear in the output list.


If clock=TRUE, the computation time is printed. Useful when nperm is large.


Calculations may be carried out in two ways, depending on the selected method.

Community composition data can be log-transformed prior to analysis with the chord distance; see Legendre and Borcard (submitted). The log(y+1) transformation (log1p function of base) reduces the asymmetry of the species distributions. The chord-log distance, readily available among the methods of the beta.div function, is the chord distance computed on log(y+1)-transformed data. This combined transformation is meaningful for community composition data because the log is one of the transformations in the Box-Cox series, corresponding to exponent 0; see Legendre and Legendre (2012, Section 1.5.6). Exponent 1 (no transformation of the data) followed by the chord transformation and calculation of the Euclidean distance would simply produce the chord distance. Exponent 0.5 (square root) followed by the chord transformation and the Euclidean distance would produce the Hellinger distance. The chord, Hellinger and log-chord distances represent a series where the data are increasingly transformed to reduce the asymmetry of the distributions. Note that it is meaningless to subject log-transformed community compostion data to the "profiles", "hellinger", or "chisquare" distances available in this function.

The Jaccard, S<c3><b8>rensen and Ochiai coefficients are the binary forms of 10 of the 12 dissimilarity coefficients (including the Ru<c5><be>i<c4><8d>ka index) that are suitable for beta diversity assessment. The equivalences are described in Legendre and De C<c3><a1>ceres (2013, Table 1). These popular coefficients can be computed directly using function beta.div without going to the trouble of applying the quantitative forms of these coefficients to data reduced to presence-absence form. beta.div produces the dissimilarity matrix in the form sqrt(D), which is Euclidean. Hence for these three coefficients, function beta.div should be used with option sqrt.D=FALSE.

Species contributions to beta diversity (SCBD indices for the species) are computed for untransformed or transformed raw data, but they cannot be computed from dissimilarity matrices.

Local contributions to beta diversity (LCBD indices) represent the degree of uniqueness of the sites in terms of their species compositions. They can be computed in all cases: raw (not recommended) or transformed data, as well as dissimilarity matrices. See Legendre and De C<c3><a1>ceres (2013) for details. LCBD indices are tested for significance by random, independent permutations within the columns of Y. This permutation method tests H0 that the species are distributed at random, independently of one another, among the sites, while preserving the species abundance distributions in the observed data. See Legendre and De C<c3><a1>ceres (2013) for discussion.

This version of beta.div calls computer code written in C to speed up computation, especially for the permutation tests of the LCBD indices.


A list containing the following results:

When all sites contain a different set of species with no species in common, the maximum value that BDtotal can take depends on the method used in the calculation.

See Legendre & De Caceres (2013, p. 957<e2><80><93>958), Table 2 and section Maximum value of BD.
For two sites only, the LCBD results are not interesting. With all coefficients, the two LCBD indices are equal to 0.5. The two associated p-values are 1 because LCBD is 0.5 for all columnwise permutations of the data.
The calculation is aborted when Y only contains two identical rows of data. In that case, SStotal and BDtotal are 0 and the LCBD indices cannot be computed (value NaN).


Pierre Legendre [email protected]


Chao, A., R. L. Chazdon, R. K. Colwell and T. J. Shen. 2006. Abundance-based similarity indices and their estimation when there are unseen species in samples. Biometrics 62: 361<e2><80><93>371.

Legendre, P. 2014. Interpreting the replacement and richness difference components of beta diversity. Global Ecology and Biogeography 23: 1324-1334.

Legendre, P. and D. Borcard. (Submitted). Box-Cox-chord transformations for community composition data prior to beta diversity analysis.

Legendre, P. and M. De C<c3><a1>ceres. 2013. Beta diversity as the variance of community data: dissimilarity coefficients and partitioning. Ecology Letters 16: 951-963.

Legendre, P. and E. D. Gallagher, E.D. 2001. Ecologically meaningful transformations for ordination of species data. Oecologia 129: 271<e2><80><93>280.

Legendre, P. and Legendre, L. 2012. Numerical Ecology. 3rd English edition. Elsevier Science BV, Amsterdam.


if(require("vegan", quietly = TRUE) & require("adegraphics", quietly = TRUE)){
res = beta.div(mite, "hellinger", nperm=999)

# Plot a map of the LCBD indices using the Cartesian coordinates
s.value(mite.xy, res$LCBD, symbol = "circle", col = c("white", "brown"), main="Map of mite LCBD")

### Example using the mite abundance data and the percentage difference dissimilarity
res = beta.div(mite, "percentdiff", nperm=999, clock=TRUE)

# Plot a map of the LCBD indices
signif = which(res$p.LCBD <= 0.05)	# Which are the significant LCBD indices?
nonsignif = which(res$p.LCBD > 0.05)	# Which are the non-significant LCBD indices?
g1 <- s.value(mite.xy[signif,], res$LCBD[signif], ppoint.alpha = 0.5, plegend.drawKey = FALSE,
 symbol = "circle", col = c("white", "red"), main="Map of mite LCBD (red = significant indices)")
g2 <- s.value(mite.xy[nonsignif,], res$LCBD[nonsignif], ppoint.alpha = 0.5,
symbol = "circle", col = c("white", "blue"))

adespatial documentation built on Jan. 20, 2018, 9:22 a.m.